👁️ VisionAI ProDetect — Report
Powered by Proximaed.com
🧠 Detection Model
Switch model → re-upload🏆 Open-Source Credits & Attributions
VisionAI ProDetect is built on top of world-class open-source research. Full credit and recognition goes to these teams.
🤗 Transformers.js v3 Engine
HuggingFace · github.com/huggingface/transformers.js
Enables state-of-the-art ML models to run entirely in-browser with WebAssembly & WebGPU. Joshua Lochner (Xenova) — primary maintainer.
📐 DETR — End-to-End Object Detection 42% mAP
Facebook AI Research (FAIR) · arxiv.org/abs/2005.12872
Nicolas Carion, Francisco Massa et al. First transformer-based detector, eliminating anchor boxes and NMS. Trained on COCO 2017 (80 classes).
⚡ RT-DETR — Real-Time Detection Transformer 53% mAP
Baidu Inc. · arxiv.org/abs/2304.08069
Wenyu Lv, Yian Zhao et al. Outperforms all YOLO models in speed & accuracy. 114 FPS on T4 GPU at 53.0% mAP COCO val2017.
🦅 YOLOS — You Only Look at One Sequence 36% mAP
Huazhong University of Science & Technology · arxiv.org/abs/2106.00666
Yuxin Fang, Bencheng Liao et al. Pure-transformer object detector inspired by ViT. Fast, lightweight, runs fully in browser.
📊 COCO Dataset — Common Objects in Context
Microsoft Research · cocodataset.org
Tsung-Yi Lin et al. The gold-standard benchmark for object detection: 330K images, 1.5M instances across 80 object categories.
📄 html2pdf.js — PDF Export
eKoopmans · github.com/eKoopmans/html2pdf.js
Open-source library for client-side HTML-to-PDF conversion using jsPDF and html2canvas. Used for in-app report generation.
⚠️ All model weights, training code and research intellectual property belong to their respective authors and institutions. VisionAI ProDetect only uses these open-source models for inference under their respective licenses (Apache 2.0 / MIT). Full credit, recognition and gratitude goes to all contributors. · Built by Proximaed.com